Application of Fisher's ratio technique for speech feature evaluation of vowls/consonants

Feature extraction technique plays an important role in Automatic speech recognition. There are many techniques for extracting the speech features. Capability of various speech features for categorizing speech sounds differs and can be evaluated after applying classification or clustering model. As an alternative approach, a preliminary study (before classification) can be done about distinguishing capacity of a speech feature. Generally, visualization techniques are used for this purpose. In this paper, we are evaluating the capability of speech feature by computing and analyzing Fisher's ratio for the vowels and consonants. It's a well known technique but not being used widely in the context of Automatic speech recognition. The term Fisher's ratio is computed based on the centroid and variance of feature within a population and across various populations. It gives a picture about the measure of separation of objects of one class with another. The evaluation results are then compared with visualization techniques and classification accuracy.